A general framework for updating belief distributions.

Area of Science:

• Statistics and Probability
• Computational Statistics
• Decision Theory

Background:

• Traditional Bayesian inference relies on likelihood functions to update prior beliefs.
• Modeling complex data-generating mechanisms is often challenging in modern applications.
• Existing Bayesian methods struggle with parameters not directly indexing density functions.

Purpose of the Study:

• To propose a generalized Bayesian inference framework.
• To enable valid posterior updates using loss functions instead of solely likelihoods.
• To extend Bayesian inference to settings with complex or unknown data-generating processes.

Main Methods:

• Developed a framework connecting data information to functionals of interest via loss functions.
• Utilized a decision-theoretic approach with cumulative loss functions for belief updating.
• Demonstrated that the proposed method recovers traditional Bayesian updating when a likelihood is known.

Main Results:

• The proposed framework provides coherent subjective inference in more general settings than traditional methods.
• It successfully handles cases where parameters of interest do not directly define density functions.
• The approach offers a unified perspective, connecting loss-based inference with likelihood-based Bayesian inference.

Conclusions:

• The loss function-based framework offers a powerful and flexible generalization of Bayesian inference.
• This approach addresses limitations of traditional methods in complex, high-dimensional, or ill-specified models.
• It opens new avenues for Bayesian analysis in diverse scientific and applied domains.